Update: I have written a paper about selective applicative functors, and it completely supersedes this blog post (it also uses a slightly different notation). You should read the paper instead.
I often need a Haskell abstraction that supports conditions (like Monad) yet can still be statically analysed (like Applicative). In such cases people typically point to the Arrow class, more specifically ArrowChoice, but when I look it up, I find several type classes and a dozen of methods. Impressive, categorical but also quite heavy. Is there a more lightweight approach? In this blog post I’ll explore what I call selective applicative functors, which extend the Applicative type class with a single method that makes it possible to be selective about effects.
Please meet Selective:
class Applicative f => Selective f where handle :: f (Either a b) -> f (a -> b) -> f b
Think of handle as a selective function application: you apply a handler function of type a → b when given a value of type Left a, but can skip the handler (along with its effects) in the case of Right b. Intuitively, handle allows you to efficiently handle errors, i.e. perform the error-handling effects only when needed.